Meta announced Friday that it will use millions of Amazon Web Services (AWS) Graviton processors to run its artificial‑intelligence applications. The Graviton line, built on an ARM architecture, is a central processing unit rather than a graphics processing unit, a distinction that matters as AI workloads evolve beyond the training phase.
While graphics processing units remain the hardware of choice for training large models, the post‑training stage—often called inference—requires different capabilities. AI agents that perform real‑time reasoning, generate code, conduct searches or manage multi‑step tasks place heavy demands on compute resources. AWS says its latest Graviton generation is engineered specifically for those demands.
The deal shifts a sizable portion of Meta’s cloud spend back to Amazon, away from rivals such as Google Cloud. Last August, Meta signed a six‑year, $10 billion agreement with Google Cloud, even though the company had primarily used AWS and also tapped Microsoft Azure. By announcing the new partnership as the Google Cloud Next conference wrapped up, Amazon sent a clear signal to its competitor.
Amazon’s chip strategy extends beyond CPUs. The company also offers the Trainium accelerator, a custom AI GPU that handles both training and inference workloads. Earlier this month, Anthropic committed to spend $100 billion over a decade on AWS, focusing on Trainium, while Amazon pledged an additional $5 billion investment in the Claude‑maker, bringing its total stake to $13 billion.
Meta’s embrace of Graviton CPUs gives Amazon a high‑profile customer to showcase its home‑grown silicon. The chips compete directly with Nvidia’s Vera CPU, another ARM‑based processor aimed at AI agentic workloads. Unlike Nvidia, which sells its chips and AI systems to enterprises and cloud providers, AWS only offers access to its silicon through its cloud platform.
In his annual shareholder letter, AWS chief Andy Jassy criticized Nvidia and Intel, arguing that enterprises seek better price‑performance ratios for AI. He pledged to win business on that basis, putting pressure on Amazon’s internal chip‑building team to deliver results.
The partnership reflects a broader trend: as AI models mature, the industry is looking for specialized, cost‑effective hardware to power the next generation of intelligent applications. Meta’s decision to lean on Amazon’s Graviton CPUs marks a significant vote of confidence in AWS’s ability to meet those emerging needs.
Este artigo foi escrito com a assistência de IA.
News Factory SEO ajuda você a automatizar conteúdo de notícias para o seu site.